With the general understanding of multi-threaded rendering in place, we can now look at the Scaleform 4.0 render design diagram.

Figure 1: Render design

The diagram illustrates renderer subsystem interactions when running with two threads. On the left side, the Movie and MovieDef objects represent Scaleform movie instance and its internal shared data, respectively. The Render Context is a front-end to the rendering subsystem that manages the Render Tree; it is responsible for 2D scene graph snapshots and change lists accessible by the render tree. In Scaleform, render context is contained within the Movie and doesn’t need to be accessed by end users.

On the right side, the render thread maintains Render::HAL. Render::HAL is the core of the render engine; it contains the Display method used to output movie snapshots to screen. Render::HAL stands for a “hardware abstraction layer” and is an abstract class having different implementation for each platform. Render HAL is responsible for executing graphics commands and managing resources such as vertex and texture buffers.

As seen from the diagram, the majority of the rendering systems are maintained and executed on the render thread. Specifically, the render thread is responsible for all of the following:

• Maintaining a sorted, cached version of the render tree and updating it when changes arrive from a new snapshot.
• Maintaining the Font Cache and updating it when new glyphs need to be rendered.
• Maintaining the Mesh Cache, which contains vertex and index buffers of tessellated vector data.
• Managing a Render Queue that enables efficient buffer update/lock management and minimizes CPU stalls while rendering.
• Emitting graphics commands to the device though HAL APIs.

In most cases keeping cache management logic on the render thread improves performance, as it frees up main thread for other work-intensive tasks such as ActionScript or game AI. Keeping caches on the rendering thread also reduces synchronization overhead and memory use, as it eliminates the need for extra buffers that would be needed to copy vertex and/or glyph data between threads.